Metabolomic Data Analysis
Advanced Strategies for Metabolomic Data Analysis
Dmitry Grapov, PhD
Analysis at the Metabolomic Scale
Multivariate Analysis
variables
Samples
Multivariate Analysis
Simultaneous analysis of many variables
Visualization Clustering Projection Modeling Networks
Clustering
Identify
patterns
group structure
relationships
Evaluate/refine hypothesis Reduce complexity
Artist: Chuck Close
Cluster Analysis
Use the concept similarity/dissimilarity to group a collection of samples or variables Linkage
Approaches hierarchical (HCA) non-hierarchical (k-NN, k-means) distribution (mixtures models) density (DBSCAN) self organizing maps (SOM)
k-means
Distribution
Density
Hierarchical Cluster Analysis
similarity/dissimilarity defines nearness or distance
euclidean manhattan Mahalanobis non-euclidean
X Y
Y
Hierarchical Cluster Analysis
Agglomerative/linkage algorithm defines how points are grouped
single
complete centroid average
Hierarchical Cluster Analysis (cont.)
x x
Similarity
Hierarchical Cluster Analysis (cont.)
How does my metadata match my data structure?
Overview
Confirmation
Multidimensional Scaling
PLoS ONE 7(11): e48852. doi:10.1371/[Link].0048852
Projection of Data
The algorithm defines the position of the light source
Principal Components Analysis (PCA) unsupervised maximize variance (X) Partial Least Squares Projection to Latent Structures (PLS) supervised maximize covariance (Y ~ X)
PCA: Goals
Principal Components (PCs) non-supervised projection of the data which maximize variance explained
Results
1. eigenvalues = variance explained
2. scores = new coordinates for samples (rows)
3. loadings = linear combination of original variables
James X. Li, 2009, VisuMap Tech.
Interpreting PCA Results
Variance explained (eigenvalues)
Row (sample) scores and column (variable) loadings
PCA Example
glucose
219021
*no scaling or centering
How are scores and loadings related?
Centering and Scaling
van den Berg RA, Hoefsloot HC, Westerhuis JA, Smilde AK, van der Werf MJ (2006) Centering, scaling, and transformations: improving the biological information content of metabolomics data. BMC Genomics 7: 142.
Data scaling is very important!
glucose (clinical)
glucose (GC/TOF)
219021
*autoscaling (unit variance and centered)
Use PLS to test a hypothesis
time = 0 120 min.
Loadings on the first latent variable (x-axis) can be used to interpret the multivariate changes in metabolites which are correlated with time
Modeling multifactorial relationships
~two-way ANOVA dynamic changes among groups
goodness of the model is all about the perspective
Determine in-sample (Q2) and outof-sample error (RMSEP) and compare to a random model permutation tests
training/testing
Biological Interpretation
Projection or mapping of analysis results into a biological context.
Visualization Enrichment Networks
biochemical structural spectral empirical
Ingredients for Network Analysis
1. Determine connections
biochemical (substrate/product) chemical similarity spectral similarity empirical dependency (correlation)
magnitude importance direction relationships
2. Determine vertex properties
Making Connections Based on Biochemistry
Organism specific biochemical relationships and information Multiple organism DBs KEGG BioCyc
Reactome
Human HMDB SMPDB
Biochemical Networks
Making Connections Based on structural similarity
Use structure to generate molecular fingerprint Calculate similarities between metabolites based on fingerprint PubChem service for similarity calculations
[Link]
BMC Bioinformatics 2012, 13:99 doi:10.1186/1471-2105-13-99
online tools
[Link]
Structural Similarity Network
Making Connections Based on spectral similarity
Connect molecules based on EI or MS/MS spectral similarity Useful for linking annotated analytes (known) to unknown
Watrous J et al. PNAS 2012;109:E1743-E1752
Spectral Similarity Network
Watrous J et al. PNAS 2012;109:E1743-E1752
Making connections based on empirical relationships
Connect molecules based on strength of correlation or partialcorrelation
Treatment Effects Network
Metabolites Shape = increase/decrease Size = importance (loading) Color = correlation
Connections red = Biochemical relationships violet = Structural similarity
Summary
Multivariate analysis is useful for: Visualization Exploration and overview Complexity reduction Identification of multidimensional relationships and trends Mapping to networks Generating holistic summaries of findings
Resource
Mapping tools (review)
Brief Bioinform (2012) doi: 10.1093/bib/bbs055
Tutorials and Examples
[Link] [Link]
Chemical Translations Services CTS: [Link] R-interface: [Link] CIR: [Link] R-interface: [Link]
